github/DB-GPT
1
0
Fork 0
mirror of https://github.com/csunny/DB-GPT.git synced 2025-08-20 01:07:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix pylint

Browse Source
This commit is contained in:
FangYin Cheng 2023-07-13 18:11:41 +08:00
parent 189ac995ec
commit accce56d49
2 changed files with 23 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
pilot/model/adapter.py
View File

@ -73,6 +73,7 @@ class VicunaLLMAdapater(BaseLLMAdaper):
)
return model, tokenizer
def auto_configure_device_map(num_gpus):
"""handling multi gpu calls"""
# transformer.word_embeddings occupying 1 floors
@ -88,11 +89,11 @@ def auto_configure_device_map(num_gpus):
# If transformer. word_ If embeddings. device and model. device are different, it will cause a RuntimeError
# Therefore, here we will transform. word_ Embeddings, transformer. final_ Layernorm, lm_ Put all the heads on the first card
device_map = {
'transformer.embedding.word_embeddings': 0,
'transformer.encoder.final_layernorm': 0,
'transformer.output_layer': 0,
'transformer.rotary_pos_emb': 0,
'lm_head': 0
"transformer.embedding.word_embeddings": 0,
"transformer.encoder.final_layernorm": 0,
"transformer.output_layer": 0,
"transformer.rotary_pos_emb": 0,
"lm_head": 0,
}
used = 2
@ -102,7 +103,7 @@ def auto_configure_device_map(num_gpus):
gpu_target += 1
used = 0
assert gpu_target < num_gpus
device_map[f'transformer.encoder.layers.{i}'] = gpu_target
device_map[f"transformer.encoder.layers.{i}"] = gpu_target
used += 1
return device_map
@ -114,7 +115,13 @@ class ChatGLMAdapater(BaseLLMAdaper):
def match(self, model_path: str):
return "chatglm" in model_path
def loader(self, model_path: str, from_pretrained_kwargs: dict, device_map=None, num_gpus=CFG.NUM_GPUS):
def loader(
self,
model_path: str,
from_pretrained_kwargs: dict,
device_map=None,
num_gpus=CFG.NUM_GPUS,
):
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
if DEVICE != "cuda":
@ -125,10 +132,8 @@ class ChatGLMAdapater(BaseLLMAdaper):
else:
model = (
AutoModel.from_pretrained(
model_path, trust_remote_code=True,
**from_pretrained_kwargs
)
.half()
model_path, trust_remote_code=True, **from_pretrained_kwargs
).half()
# .cuda()
)
from accelerate import dispatch_model